[10:54:09] <mszabo>	 https://gerrit.wikimedia.org/r/c/mediawiki/core/+/1017264
[10:54:22] <mszabo>	 I'm not sure of the exact impact here, because the data on https://grafana.wikimedia.org/d/lqE4lcGWz/wanobjectcache-key-group?orgId=1&var-kClass=SqlBlobStore_blob is a bit odd
[10:54:45] <mszabo>	 I'd expect hit.refresh ~= renew.compute, yet there is always a notable difference
[10:58:48] <mszabo>	 as far as I see, hit.refresh increments when an async update was fired off to do the preemptive refresh, which would then increment renew.compute
[10:58:54] <mszabo>	 so not sure where the discrepancy is coming from
[18:14:32] <Krinkle>	 mszabo: good question. I can see why it would be lower, eg some might turn into "busy", "miss", or otherwise lost.
[18:14:41] <Krinkle>	 But 10x higher is surprising at glance
[18:15:10] <Krinkle>	 I wonder if it's happening before instead of after pcTTL handling
[18:15:23] <Krinkle>	 If before that would explain since we don't increment hits more than once
[18:15:48] <Krinkle>	 Afaik that code is really early though so unlikely